Am 12.01.2018 01:31, schrieb Dieter Nützel:
Tested-by: Dieter Nützel <die...@nuetzel-hh.de>
This offer an additional 1-1,5% (~20 fps) boost with 'smoketest'.
6290 presents in 5.00045 seconds (FPS: 1257.89)
6291 presents in 5.00041 seconds (FPS: 1258.1)
6293 presents in 5.00065 seconds (FPS: 1258.44)
6289 presents in 5.0006 seconds (FPS: 1257.65)
But _only_ with 'ondemand' governor!!!
cpupower frequency-set -g ondemand => GOOD
cpupower frequency-set -g performance => BAD
cpupower frequency-set -g schedutil => BAD
performance: 100-110 fps _lower_
schedutil: 110-140 fps _lower_
Hello Dave,
if you refer to this part,
I'm not sure this was much of a win, and if it makes things worse in
some
cases, then it needs a lot more investigation, so probably consider it
dead
for now.
Dave.
it was _only_ additional finding on my side.
Boost is there, I'm running daily with it.
'New' finding for me was, that 'performance' and 'schedutil' is slower
as 'ondemand' for Vulkan (only radv tested) opposed to OpenGL, where
'performance' is the clear winner (in most games much more min and avg
plus max fps).
With 'cpupower frequency-set -g performance' I get _more_ noise (bus?)
from my system during 'smoketest'. (Apart from this patch.) And it is
somewhat slower.
Your call.
Greetings,
Dieter
Am 11.01.2018 04:53, schrieb Dave Airlie:
From: Dave Airlie <airl...@redhat.com>
This should make the merging of cmd buffers less CPU intensive,
note I said *should* :)
---
src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c | 47
++++++++++++---------------
1 file changed, 20 insertions(+), 27 deletions(-)
diff --git a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c
b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c
index 0ee56f91447..9a39d237ae8 100644
--- a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c
+++ b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c
@@ -33,6 +33,7 @@
#include "radv_amdgpu_bo.h"
#include "sid.h"
+#include "util/hash_table.h"
enum {
VIRTUAL_BUFFER_HASH_TABLE_SIZE = 1024
@@ -584,6 +585,9 @@ static int radv_amdgpu_create_bo_list(struct
radv_amdgpu_winsys *ws,
priorities[0] = 8;
}
+ struct hash_table *ht = _mesa_hash_table_create(NULL,
_mesa_hash_pointer,
+
_mesa_key_pointer_equal);
+
for (unsigned i = 0; i < count + !!extra_cs; ++i) {
struct radv_amdgpu_cs *cs;
@@ -595,50 +599,39 @@ static int radv_amdgpu_create_bo_list(struct
radv_amdgpu_winsys *ws,
if (!cs->num_buffers)
continue;
- if (unique_bo_count == 0) {
- memcpy(handles, cs->handles, cs->num_buffers *
sizeof(amdgpu_bo_handle));
- memcpy(priorities, cs->priorities, cs->num_buffers *
sizeof(uint8_t));
- unique_bo_count = cs->num_buffers;
- continue;
- }
- int unique_bo_so_far = unique_bo_count;
for (unsigned j = 0; j < cs->num_buffers; ++j) {
- bool found = false;
- for (unsigned k = 0; k < unique_bo_so_far; ++k)
{
- if (handles[k] == cs->handles[j]) {
- found = true;
- priorities[k] =
MAX2(priorities[k],
-
cs->priorities[j]);
- break;
- }
- }
- if (!found) {
+ struct hash_entry *entry =
_mesa_hash_table_search(ht, (void
*)cs->handles[j]);
+ if (!entry) {
+ _mesa_hash_table_insert(ht, (void
*)cs->handles[j], (void
*)(uintptr_t)unique_bo_count);
handles[unique_bo_count] =
cs->handles[j];
priorities[unique_bo_count] =
cs->priorities[j];
++unique_bo_count;
+ } else {
+ int bo_idx = (uint32_t)(unsigned
long)entry->data;
+ priorities[bo_idx] =
MAX2(priorities[bo_idx],
+
cs->priorities[j]);
}
}
for (unsigned j = 0; j < cs->num_virtual_buffers; ++j) {
struct radv_amdgpu_winsys_bo *virtual_bo =
radv_amdgpu_winsys_bo(cs->virtual_buffers[j]);
for(unsigned k = 0; k < virtual_bo->bo_count;
++k) {
struct radv_amdgpu_winsys_bo *bo =
virtual_bo->bos[k];
- bool found = false;
- for (unsigned m = 0; m <
unique_bo_count; ++m) {
- if (handles[m] == bo->bo) {
- found = true;
- priorities[m] =
MAX2(priorities[m],
-
cs->virtual_buffer_priorities[j]);
- break;
- }
- }
- if (!found) {
+
+ struct hash_entry *entry = _mesa_hash_table_search(ht, (void
*)bo->bo);
+ if (!entry) {
+ _mesa_hash_table_insert(ht, (void
*)bo->bo, (void
*)(uintptr_t)unique_bo_count);
handles[unique_bo_count] =
bo->bo;
priorities[unique_bo_count] =
cs->virtual_buffer_priorities[j];
++unique_bo_count;
+ } else {
+ int bo_idx = (uint32_t)(unsigned
long)entry->data;
+ priorities[bo_idx] =
MAX2(priorities[bo_idx],
+
cs->virtual_buffer_priorities[j]);
}
}
}
}
+ _mesa_hash_table_destroy(ht, NULL);
if (unique_bo_count > 0) {
r = amdgpu_bo_list_create(ws->dev, unique_bo_count,
handles,
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
